Calibrated Lazy Associative Classification

Classi cation is an important problem in data mining. Given an example x and a class c, a classi er usually works by estimating the probability of x being member of c (i.e., membership probability). Well calibrated classi ers are those able to provide accurate estimates of class membership probabilities, that is, the estimated probability p̂(c|x) is close to p(c|p̂(c|x)), which is the true, empirical probability of x being member of c given that the probability estimated by the classi er is p̂(c|x). Calibration is not a necessary property for producing accurate classi ers, and thus, most of the research has focused on direct accuracy maximization strategies (i.e., maximum margin) rather than on calibration. However, non-calibrated classi ers are problematic in applications where the reliability associated with a prediction must be taken into account (i.e., cost-sensitive classi cation, cautious classi cation etc.). In these applications, a sensible use of the classi er must be based on the reliability of its predictions, and thus, the classi er must be well calibrated. In this paper we show that lazy associative classi ers (LAC) are accurate, and well calibrated using a well known, sound, entropy-minimization method. We explore important applications where such characteristics (i.e., accuracy and calibration) are relevant, and we demonstrate empirically that LAC drastically outperforms other classi ers, such as SVMs, Naive Bayes, and Decision Trees (even after these classi ers are calibrated by speci c methods). Additional highlights of LAC include the ability to incorporate reliable predictions for improving training, and the ability to refrain from doubtful predictions.